This study investigates the atmospheric fine particulate matter (PM 2.5 ) issue caused by the multi-effect of complicated sources, terrain, and meteorology at Southern Taiwan. Three sampling stations represent an urban, a rural, and a coastal sites near an urban-industrial complex. The atmospheric PM 2.5 were measured during a pollution episode from November to February, when the reference samples were collected in April. The sample was collected with a constant-flow sampler with the Federal Reference Method performance. After determining the PM 2.5 mass, their chemical compositions of ions, metals, and carbons were analyzed for the different properties caused by the multi-factors. The chemical mass balance (CMB) model was employed to evaluate the emission contributions. Additionally, an inverse trajectory model is used to analyze the pollutant transport and support the CMB results. The air-pollution episodes occurred within the winter to spring. The PM2.5 were composed of 51-69% ions, 18-31% carbonaceous species, and 1.5-3.0% metals. The SO 4 2-, NH 4 + , and NO 3contributed 92-96% of the ions. Most of the organic/elemental carbon ratios were low, suggesting more primary carbon emissions. The metal contents were minor and dominated by Fe and Zn. The CMB model indicated the PM 2.5 were dominated by 24% secondary SO 4 2-, 14.7% traffics, 8.3% petrochemical emissions, 6.8% soil dust, and 4.5% sintering plant emission. For non-episode days, the PM 2.5 were contributed by 34.9% traffics, 30% the secondary SO 4 2-, 10.3% secondary NO 3 -, and 6.8% soil dust. Nevertheless, the frequent sea-land breeze might lead to more powerful wind eddies and bring the primary PM 2.5 and the aerosol precursors from the emission areas. Consequently, the uncontrollable meteorological changings would lead to the pollution issues at the lowly convective area. Therefore, the averaging emissions of the PM 2.5 and precursors should be lowered; meanwhile, the rapid controls of the primary emissions are suggested when the high-level PM 2.5 are forecasted.